Nix/src/libutil/args.cc

509 lines
16 KiB
C++
Raw Normal View History

#include "args.hh"
Overhaul completions, redo #6693 (#8131) As I complained in https://github.com/NixOS/nix/pull/6784#issuecomment-1421777030 (a comment on the wrong PR, sorry again!), #6693 introduced a second completions mechanism to fix a bug. Having two completion mechanisms isn't so nice. As @thufschmitt also pointed out, it was a bummer to go from `FlakeRef` to `std::string` when collecting flake refs. Now it is `FlakeRefs` again. The underlying issue that sought to work around was that completion of arguments not at the end can still benefit from the information from latter arguments. To fix this better, we rip out that change and simply defer all completion processing until after all the (regular, already-complete) arguments have been passed. In addition, I noticed the original completion logic used some global variables. I do not like global variables, because even if they save lines of code, they also obfuscate the architecture of the code. I got rid of them moved them to a new `RootArgs` class, which now has `parseCmdline` instead of `Args`. The idea is that we have many argument parsers from subcommands and what-not, but only one root args that owns the other per actual parsing invocation. The state that was global is now part of the root args instead. This did, admittedly, add a bunch of new code. And I do feel bad about that. So I went and added a lot of API docs to try to at least make the current state of things clear to the next person. -- This is needed for RFC 134 (tracking issue #7868). It was very hard to modularize `Installable` parsing when there were two completion arguments. I wouldn't go as far as to say it is *easy* now, but at least it is less hard (and the completions test finally passed). Co-authored-by: Valentin Gagarin <valentin.gagarin@tweag.io>
2023-10-23 15:03:11 +02:00
#include "args/root.hh"
#include "hash.hh"
#include "environment-variables.hh"
#include "signals.hh"
#include "users.hh"
#include "json-utils.hh"
2021-09-01 08:19:51 +02:00
#include <fstream>
#include <string>
#include <regex>
2020-05-10 21:35:07 +02:00
#include <glob.h>
namespace nix {
2020-05-04 22:40:19 +02:00
void Args::addFlag(Flag && flag_)
{
2020-05-04 22:40:19 +02:00
auto flag = std::make_shared<Flag>(std::move(flag_));
if (flag->handler.arity != ArityAny)
assert(flag->handler.arity == flag->labels.size());
assert(flag->longName != "");
2020-05-04 22:40:19 +02:00
longFlags[flag->longName] = flag;
for (auto & alias : flag->aliases)
longFlags[alias] = flag;
2020-05-04 22:40:19 +02:00
if (flag->shortName) shortFlags[flag->shortName] = flag;
}
void Args::removeFlag(const std::string & longName)
{
auto flag = longFlags.find(longName);
assert(flag != longFlags.end());
if (flag->second->shortName) shortFlags.erase(flag->second->shortName);
longFlags.erase(flag);
}
Overhaul completions, redo #6693 (#8131) As I complained in https://github.com/NixOS/nix/pull/6784#issuecomment-1421777030 (a comment on the wrong PR, sorry again!), #6693 introduced a second completions mechanism to fix a bug. Having two completion mechanisms isn't so nice. As @thufschmitt also pointed out, it was a bummer to go from `FlakeRef` to `std::string` when collecting flake refs. Now it is `FlakeRefs` again. The underlying issue that sought to work around was that completion of arguments not at the end can still benefit from the information from latter arguments. To fix this better, we rip out that change and simply defer all completion processing until after all the (regular, already-complete) arguments have been passed. In addition, I noticed the original completion logic used some global variables. I do not like global variables, because even if they save lines of code, they also obfuscate the architecture of the code. I got rid of them moved them to a new `RootArgs` class, which now has `parseCmdline` instead of `Args`. The idea is that we have many argument parsers from subcommands and what-not, but only one root args that owns the other per actual parsing invocation. The state that was global is now part of the root args instead. This did, admittedly, add a bunch of new code. And I do feel bad about that. So I went and added a lot of API docs to try to at least make the current state of things clear to the next person. -- This is needed for RFC 134 (tracking issue #7868). It was very hard to modularize `Installable` parsing when there were two completion arguments. I wouldn't go as far as to say it is *easy* now, but at least it is less hard (and the completions test finally passed). Co-authored-by: Valentin Gagarin <valentin.gagarin@tweag.io>
2023-10-23 15:03:11 +02:00
void Completions::setType(AddCompletions::Type t)
{
type = t;
}
void Completions::add(std::string completion, std::string description)
{
2023-02-10 22:17:09 +01:00
description = trim(description);
// ellipsize overflowing content on the back of the description
auto end_index = description.find_first_of(".\n");
if (end_index != std::string::npos) {
auto needs_ellipsis = end_index != description.size() - 1;
description.resize(end_index);
if (needs_ellipsis)
description.append(" [...]");
}
Overhaul completions, redo #6693 (#8131) As I complained in https://github.com/NixOS/nix/pull/6784#issuecomment-1421777030 (a comment on the wrong PR, sorry again!), #6693 introduced a second completions mechanism to fix a bug. Having two completion mechanisms isn't so nice. As @thufschmitt also pointed out, it was a bummer to go from `FlakeRef` to `std::string` when collecting flake refs. Now it is `FlakeRefs` again. The underlying issue that sought to work around was that completion of arguments not at the end can still benefit from the information from latter arguments. To fix this better, we rip out that change and simply defer all completion processing until after all the (regular, already-complete) arguments have been passed. In addition, I noticed the original completion logic used some global variables. I do not like global variables, because even if they save lines of code, they also obfuscate the architecture of the code. I got rid of them moved them to a new `RootArgs` class, which now has `parseCmdline` instead of `Args`. The idea is that we have many argument parsers from subcommands and what-not, but only one root args that owns the other per actual parsing invocation. The state that was global is now part of the root args instead. This did, admittedly, add a bunch of new code. And I do feel bad about that. So I went and added a lot of API docs to try to at least make the current state of things clear to the next person. -- This is needed for RFC 134 (tracking issue #7868). It was very hard to modularize `Installable` parsing when there were two completion arguments. I wouldn't go as far as to say it is *easy* now, but at least it is less hard (and the completions test finally passed). Co-authored-by: Valentin Gagarin <valentin.gagarin@tweag.io>
2023-10-23 15:03:11 +02:00
completions.insert(Completion {
.completion = completion,
.description = description
});
}
bool Completion::operator<(const Completion & other) const
{ return completion < other.completion || (completion == other.completion && description < other.description); }
2020-05-10 20:32:21 +02:00
std::string completionMarker = "___COMPLETE___";
Overhaul completions, redo #6693 (#8131) As I complained in https://github.com/NixOS/nix/pull/6784#issuecomment-1421777030 (a comment on the wrong PR, sorry again!), #6693 introduced a second completions mechanism to fix a bug. Having two completion mechanisms isn't so nice. As @thufschmitt also pointed out, it was a bummer to go from `FlakeRef` to `std::string` when collecting flake refs. Now it is `FlakeRefs` again. The underlying issue that sought to work around was that completion of arguments not at the end can still benefit from the information from latter arguments. To fix this better, we rip out that change and simply defer all completion processing until after all the (regular, already-complete) arguments have been passed. In addition, I noticed the original completion logic used some global variables. I do not like global variables, because even if they save lines of code, they also obfuscate the architecture of the code. I got rid of them moved them to a new `RootArgs` class, which now has `parseCmdline` instead of `Args`. The idea is that we have many argument parsers from subcommands and what-not, but only one root args that owns the other per actual parsing invocation. The state that was global is now part of the root args instead. This did, admittedly, add a bunch of new code. And I do feel bad about that. So I went and added a lot of API docs to try to at least make the current state of things clear to the next person. -- This is needed for RFC 134 (tracking issue #7868). It was very hard to modularize `Installable` parsing when there were two completion arguments. I wouldn't go as far as to say it is *easy* now, but at least it is less hard (and the completions test finally passed). Co-authored-by: Valentin Gagarin <valentin.gagarin@tweag.io>
2023-10-23 15:03:11 +02:00
RootArgs & Args::getRoot()
{
Args * p = this;
while (p->parent)
p = p->parent;
auto * res = dynamic_cast<RootArgs *>(p);
assert(res);
return *res;
}
std::optional<std::string> RootArgs::needsCompletion(std::string_view s)
2020-05-10 20:32:21 +02:00
{
if (!completions) return {};
auto i = s.find(completionMarker);
if (i != std::string::npos)
return std::string(s.begin(), i);
return {};
}
Overhaul completions, redo #6693 (#8131) As I complained in https://github.com/NixOS/nix/pull/6784#issuecomment-1421777030 (a comment on the wrong PR, sorry again!), #6693 introduced a second completions mechanism to fix a bug. Having two completion mechanisms isn't so nice. As @thufschmitt also pointed out, it was a bummer to go from `FlakeRef` to `std::string` when collecting flake refs. Now it is `FlakeRefs` again. The underlying issue that sought to work around was that completion of arguments not at the end can still benefit from the information from latter arguments. To fix this better, we rip out that change and simply defer all completion processing until after all the (regular, already-complete) arguments have been passed. In addition, I noticed the original completion logic used some global variables. I do not like global variables, because even if they save lines of code, they also obfuscate the architecture of the code. I got rid of them moved them to a new `RootArgs` class, which now has `parseCmdline` instead of `Args`. The idea is that we have many argument parsers from subcommands and what-not, but only one root args that owns the other per actual parsing invocation. The state that was global is now part of the root args instead. This did, admittedly, add a bunch of new code. And I do feel bad about that. So I went and added a lot of API docs to try to at least make the current state of things clear to the next person. -- This is needed for RFC 134 (tracking issue #7868). It was very hard to modularize `Installable` parsing when there were two completion arguments. I wouldn't go as far as to say it is *easy* now, but at least it is less hard (and the completions test finally passed). Co-authored-by: Valentin Gagarin <valentin.gagarin@tweag.io>
2023-10-23 15:03:11 +02:00
void RootArgs::parseCmdline(const Strings & _cmdline)
{
// Default via 5.1.2.2.1 in C standard
Args::parseCmdline("", _cmdline);
}
void Args::parseCmdline(const std::string & programName, const Strings & _cmdline)
{
Strings pendingArgs;
bool dashDash = false;
Strings cmdline(_cmdline);
2020-05-10 20:32:21 +02:00
if (auto s = getEnv("NIX_GET_COMPLETIONS")) {
size_t n = std::stoi(*s);
assert(n > 0 && n <= cmdline.size());
*std::next(cmdline.begin(), n - 1) += completionMarker;
Overhaul completions, redo #6693 (#8131) As I complained in https://github.com/NixOS/nix/pull/6784#issuecomment-1421777030 (a comment on the wrong PR, sorry again!), #6693 introduced a second completions mechanism to fix a bug. Having two completion mechanisms isn't so nice. As @thufschmitt also pointed out, it was a bummer to go from `FlakeRef` to `std::string` when collecting flake refs. Now it is `FlakeRefs` again. The underlying issue that sought to work around was that completion of arguments not at the end can still benefit from the information from latter arguments. To fix this better, we rip out that change and simply defer all completion processing until after all the (regular, already-complete) arguments have been passed. In addition, I noticed the original completion logic used some global variables. I do not like global variables, because even if they save lines of code, they also obfuscate the architecture of the code. I got rid of them moved them to a new `RootArgs` class, which now has `parseCmdline` instead of `Args`. The idea is that we have many argument parsers from subcommands and what-not, but only one root args that owns the other per actual parsing invocation. The state that was global is now part of the root args instead. This did, admittedly, add a bunch of new code. And I do feel bad about that. So I went and added a lot of API docs to try to at least make the current state of things clear to the next person. -- This is needed for RFC 134 (tracking issue #7868). It was very hard to modularize `Installable` parsing when there were two completion arguments. I wouldn't go as far as to say it is *easy* now, but at least it is less hard (and the completions test finally passed). Co-authored-by: Valentin Gagarin <valentin.gagarin@tweag.io>
2023-10-23 15:03:11 +02:00
completions = std::make_shared<Completions>();
verbosity = lvlError;
2020-05-10 20:32:21 +02:00
}
bool argsSeen = false;
// Heuristic to see if we're invoked as a shebang script, namely,
// if we have at least one argument, it's the name of an
// executable file, and it starts with "#!".
Strings savedArgs;
auto isNixCommand = std::regex_search(programName, std::regex("nix$"));
if (isNixCommand && cmdline.size() > 0) {
auto script = *cmdline.begin();
try {
2021-09-01 08:19:51 +02:00
std::ifstream stream(script);
char shebang[3]={0,0,0};
stream.get(shebang,3);
if (strncmp(shebang,"#!",2) == 0){
for (auto pos = std::next(cmdline.begin()); pos != cmdline.end();pos++)
savedArgs.push_back(*pos);
cmdline.clear();
2021-09-01 08:19:51 +02:00
std::string line;
std::getline(stream,line);
std::string commentChars("#/\\%@*-");
while (std::getline(stream,line) && !line.empty() && commentChars.find(line[0]) != std::string::npos){
line = chomp(line);
std::smatch match;
if (std::regex_match(line, match, std::regex("^#!\\s*nix\\s(.*)$")))
for (const auto & word : shellwords(match[1].str()))
cmdline.push_back(word);
}
cmdline.push_back(script);
for (auto pos = savedArgs.begin(); pos != savedArgs.end();pos++)
cmdline.push_back(*pos);
}
} catch (SysError &) { }
}
for (auto pos = cmdline.begin(); pos != cmdline.end(); ) {
auto arg = *pos;
/* Expand compound dash options (i.e., `-qlf' -> `-q -l -f',
`-j3` -> `-j 3`). */
if (!dashDash && arg.length() > 2 && arg[0] == '-' && arg[1] != '-' && isalpha(arg[1])) {
*pos = (std::string) "-" + arg[1];
auto next = pos; ++next;
for (unsigned int j = 2; j < arg.length(); j++)
if (isalpha(arg[j]))
cmdline.insert(next, (std::string) "-" + arg[j]);
else {
cmdline.insert(next, std::string(arg, j));
break;
}
arg = *pos;
}
if (!dashDash && arg == "--") {
dashDash = true;
++pos;
}
else if (!dashDash && std::string(arg, 0, 1) == "-") {
if (!processFlag(pos, cmdline.end()))
throw UsageError("unrecognised flag '%1%'", arg);
}
else {
if (!argsSeen) {
argsSeen = true;
initialFlagsProcessed();
}
pos = rewriteArgs(cmdline, pos);
pendingArgs.push_back(*pos++);
if (processArgs(pendingArgs, false))
pendingArgs.clear();
}
}
processArgs(pendingArgs, true);
if (!argsSeen)
initialFlagsProcessed();
/* Now that we are done parsing, make sure that any experimental
* feature required by the flags is enabled */
for (auto & f : flagExperimentalFeatures)
experimentalFeatureSettings.require(f);
Overhaul completions, redo #6693 (#8131) As I complained in https://github.com/NixOS/nix/pull/6784#issuecomment-1421777030 (a comment on the wrong PR, sorry again!), #6693 introduced a second completions mechanism to fix a bug. Having two completion mechanisms isn't so nice. As @thufschmitt also pointed out, it was a bummer to go from `FlakeRef` to `std::string` when collecting flake refs. Now it is `FlakeRefs` again. The underlying issue that sought to work around was that completion of arguments not at the end can still benefit from the information from latter arguments. To fix this better, we rip out that change and simply defer all completion processing until after all the (regular, already-complete) arguments have been passed. In addition, I noticed the original completion logic used some global variables. I do not like global variables, because even if they save lines of code, they also obfuscate the architecture of the code. I got rid of them moved them to a new `RootArgs` class, which now has `parseCmdline` instead of `Args`. The idea is that we have many argument parsers from subcommands and what-not, but only one root args that owns the other per actual parsing invocation. The state that was global is now part of the root args instead. This did, admittedly, add a bunch of new code. And I do feel bad about that. So I went and added a lot of API docs to try to at least make the current state of things clear to the next person. -- This is needed for RFC 134 (tracking issue #7868). It was very hard to modularize `Installable` parsing when there were two completion arguments. I wouldn't go as far as to say it is *easy* now, but at least it is less hard (and the completions test finally passed). Co-authored-by: Valentin Gagarin <valentin.gagarin@tweag.io>
2023-10-23 15:03:11 +02:00
/* Now that all the other args are processed, run the deferred completions.
*/
for (auto d : deferredCompletions)
d.completer(*completions, d.n, d.prefix);
}
bool Args::processFlag(Strings::iterator & pos, Strings::iterator end)
{
assert(pos != end);
Overhaul completions, redo #6693 (#8131) As I complained in https://github.com/NixOS/nix/pull/6784#issuecomment-1421777030 (a comment on the wrong PR, sorry again!), #6693 introduced a second completions mechanism to fix a bug. Having two completion mechanisms isn't so nice. As @thufschmitt also pointed out, it was a bummer to go from `FlakeRef` to `std::string` when collecting flake refs. Now it is `FlakeRefs` again. The underlying issue that sought to work around was that completion of arguments not at the end can still benefit from the information from latter arguments. To fix this better, we rip out that change and simply defer all completion processing until after all the (regular, already-complete) arguments have been passed. In addition, I noticed the original completion logic used some global variables. I do not like global variables, because even if they save lines of code, they also obfuscate the architecture of the code. I got rid of them moved them to a new `RootArgs` class, which now has `parseCmdline` instead of `Args`. The idea is that we have many argument parsers from subcommands and what-not, but only one root args that owns the other per actual parsing invocation. The state that was global is now part of the root args instead. This did, admittedly, add a bunch of new code. And I do feel bad about that. So I went and added a lot of API docs to try to at least make the current state of things clear to the next person. -- This is needed for RFC 134 (tracking issue #7868). It was very hard to modularize `Installable` parsing when there were two completion arguments. I wouldn't go as far as to say it is *easy* now, but at least it is less hard (and the completions test finally passed). Co-authored-by: Valentin Gagarin <valentin.gagarin@tweag.io>
2023-10-23 15:03:11 +02:00
auto & rootArgs = getRoot();
auto process = [&](const std::string & name, const Flag & flag) -> bool {
++pos;
if (auto & f = flag.experimentalFeature)
Overhaul completions, redo #6693 (#8131) As I complained in https://github.com/NixOS/nix/pull/6784#issuecomment-1421777030 (a comment on the wrong PR, sorry again!), #6693 introduced a second completions mechanism to fix a bug. Having two completion mechanisms isn't so nice. As @thufschmitt also pointed out, it was a bummer to go from `FlakeRef` to `std::string` when collecting flake refs. Now it is `FlakeRefs` again. The underlying issue that sought to work around was that completion of arguments not at the end can still benefit from the information from latter arguments. To fix this better, we rip out that change and simply defer all completion processing until after all the (regular, already-complete) arguments have been passed. In addition, I noticed the original completion logic used some global variables. I do not like global variables, because even if they save lines of code, they also obfuscate the architecture of the code. I got rid of them moved them to a new `RootArgs` class, which now has `parseCmdline` instead of `Args`. The idea is that we have many argument parsers from subcommands and what-not, but only one root args that owns the other per actual parsing invocation. The state that was global is now part of the root args instead. This did, admittedly, add a bunch of new code. And I do feel bad about that. So I went and added a lot of API docs to try to at least make the current state of things clear to the next person. -- This is needed for RFC 134 (tracking issue #7868). It was very hard to modularize `Installable` parsing when there were two completion arguments. I wouldn't go as far as to say it is *easy* now, but at least it is less hard (and the completions test finally passed). Co-authored-by: Valentin Gagarin <valentin.gagarin@tweag.io>
2023-10-23 15:03:11 +02:00
rootArgs.flagExperimentalFeatures.insert(*f);
std::vector<std::string> args;
bool anyCompleted = false;
2020-05-04 22:40:19 +02:00
for (size_t n = 0 ; n < flag.handler.arity; ++n) {
if (pos == end) {
if (flag.handler.arity == ArityAny || anyCompleted) break;
throw UsageError(
"flag '%s' requires %d argument(s), but only %d were given",
name, flag.handler.arity, n);
}
Overhaul completions, redo #6693 (#8131) As I complained in https://github.com/NixOS/nix/pull/6784#issuecomment-1421777030 (a comment on the wrong PR, sorry again!), #6693 introduced a second completions mechanism to fix a bug. Having two completion mechanisms isn't so nice. As @thufschmitt also pointed out, it was a bummer to go from `FlakeRef` to `std::string` when collecting flake refs. Now it is `FlakeRefs` again. The underlying issue that sought to work around was that completion of arguments not at the end can still benefit from the information from latter arguments. To fix this better, we rip out that change and simply defer all completion processing until after all the (regular, already-complete) arguments have been passed. In addition, I noticed the original completion logic used some global variables. I do not like global variables, because even if they save lines of code, they also obfuscate the architecture of the code. I got rid of them moved them to a new `RootArgs` class, which now has `parseCmdline` instead of `Args`. The idea is that we have many argument parsers from subcommands and what-not, but only one root args that owns the other per actual parsing invocation. The state that was global is now part of the root args instead. This did, admittedly, add a bunch of new code. And I do feel bad about that. So I went and added a lot of API docs to try to at least make the current state of things clear to the next person. -- This is needed for RFC 134 (tracking issue #7868). It was very hard to modularize `Installable` parsing when there were two completion arguments. I wouldn't go as far as to say it is *easy* now, but at least it is less hard (and the completions test finally passed). Co-authored-by: Valentin Gagarin <valentin.gagarin@tweag.io>
2023-10-23 15:03:11 +02:00
if (auto prefix = rootArgs.needsCompletion(*pos)) {
anyCompleted = true;
Overhaul completions, redo #6693 (#8131) As I complained in https://github.com/NixOS/nix/pull/6784#issuecomment-1421777030 (a comment on the wrong PR, sorry again!), #6693 introduced a second completions mechanism to fix a bug. Having two completion mechanisms isn't so nice. As @thufschmitt also pointed out, it was a bummer to go from `FlakeRef` to `std::string` when collecting flake refs. Now it is `FlakeRefs` again. The underlying issue that sought to work around was that completion of arguments not at the end can still benefit from the information from latter arguments. To fix this better, we rip out that change and simply defer all completion processing until after all the (regular, already-complete) arguments have been passed. In addition, I noticed the original completion logic used some global variables. I do not like global variables, because even if they save lines of code, they also obfuscate the architecture of the code. I got rid of them moved them to a new `RootArgs` class, which now has `parseCmdline` instead of `Args`. The idea is that we have many argument parsers from subcommands and what-not, but only one root args that owns the other per actual parsing invocation. The state that was global is now part of the root args instead. This did, admittedly, add a bunch of new code. And I do feel bad about that. So I went and added a lot of API docs to try to at least make the current state of things clear to the next person. -- This is needed for RFC 134 (tracking issue #7868). It was very hard to modularize `Installable` parsing when there were two completion arguments. I wouldn't go as far as to say it is *easy* now, but at least it is less hard (and the completions test finally passed). Co-authored-by: Valentin Gagarin <valentin.gagarin@tweag.io>
2023-10-23 15:03:11 +02:00
if (flag.completer) {
rootArgs.deferredCompletions.push_back({
.completer = flag.completer,
.n = n,
.prefix = *prefix,
});
}
}
2020-05-10 21:50:32 +02:00
args.push_back(*pos++);
}
if (!anyCompleted)
flag.handler.fun(std::move(args));
return true;
};
if (std::string(*pos, 0, 2) == "--") {
Overhaul completions, redo #6693 (#8131) As I complained in https://github.com/NixOS/nix/pull/6784#issuecomment-1421777030 (a comment on the wrong PR, sorry again!), #6693 introduced a second completions mechanism to fix a bug. Having two completion mechanisms isn't so nice. As @thufschmitt also pointed out, it was a bummer to go from `FlakeRef` to `std::string` when collecting flake refs. Now it is `FlakeRefs` again. The underlying issue that sought to work around was that completion of arguments not at the end can still benefit from the information from latter arguments. To fix this better, we rip out that change and simply defer all completion processing until after all the (regular, already-complete) arguments have been passed. In addition, I noticed the original completion logic used some global variables. I do not like global variables, because even if they save lines of code, they also obfuscate the architecture of the code. I got rid of them moved them to a new `RootArgs` class, which now has `parseCmdline` instead of `Args`. The idea is that we have many argument parsers from subcommands and what-not, but only one root args that owns the other per actual parsing invocation. The state that was global is now part of the root args instead. This did, admittedly, add a bunch of new code. And I do feel bad about that. So I went and added a lot of API docs to try to at least make the current state of things clear to the next person. -- This is needed for RFC 134 (tracking issue #7868). It was very hard to modularize `Installable` parsing when there were two completion arguments. I wouldn't go as far as to say it is *easy* now, but at least it is less hard (and the completions test finally passed). Co-authored-by: Valentin Gagarin <valentin.gagarin@tweag.io>
2023-10-23 15:03:11 +02:00
if (auto prefix = rootArgs.needsCompletion(*pos)) {
2020-05-10 20:32:21 +02:00
for (auto & [name, flag] : longFlags) {
if (!hiddenCategories.count(flag->category)
&& hasPrefix(name, std::string(*prefix, 2)))
{
if (auto & f = flag->experimentalFeature)
Overhaul completions, redo #6693 (#8131) As I complained in https://github.com/NixOS/nix/pull/6784#issuecomment-1421777030 (a comment on the wrong PR, sorry again!), #6693 introduced a second completions mechanism to fix a bug. Having two completion mechanisms isn't so nice. As @thufschmitt also pointed out, it was a bummer to go from `FlakeRef` to `std::string` when collecting flake refs. Now it is `FlakeRefs` again. The underlying issue that sought to work around was that completion of arguments not at the end can still benefit from the information from latter arguments. To fix this better, we rip out that change and simply defer all completion processing until after all the (regular, already-complete) arguments have been passed. In addition, I noticed the original completion logic used some global variables. I do not like global variables, because even if they save lines of code, they also obfuscate the architecture of the code. I got rid of them moved them to a new `RootArgs` class, which now has `parseCmdline` instead of `Args`. The idea is that we have many argument parsers from subcommands and what-not, but only one root args that owns the other per actual parsing invocation. The state that was global is now part of the root args instead. This did, admittedly, add a bunch of new code. And I do feel bad about that. So I went and added a lot of API docs to try to at least make the current state of things clear to the next person. -- This is needed for RFC 134 (tracking issue #7868). It was very hard to modularize `Installable` parsing when there were two completion arguments. I wouldn't go as far as to say it is *easy* now, but at least it is less hard (and the completions test finally passed). Co-authored-by: Valentin Gagarin <valentin.gagarin@tweag.io>
2023-10-23 15:03:11 +02:00
rootArgs.flagExperimentalFeatures.insert(*f);
rootArgs.completions->add("--" + name, flag->description);
}
2020-05-10 20:32:21 +02:00
}
return false;
2020-05-10 20:32:21 +02:00
}
auto i = longFlags.find(std::string(*pos, 2));
if (i == longFlags.end()) return false;
return process("--" + i->first, *i->second);
}
if (std::string(*pos, 0, 1) == "-" && pos->size() == 2) {
auto c = (*pos)[1];
auto i = shortFlags.find(c);
if (i == shortFlags.end()) return false;
return process(std::string("-") + c, *i->second);
}
Overhaul completions, redo #6693 (#8131) As I complained in https://github.com/NixOS/nix/pull/6784#issuecomment-1421777030 (a comment on the wrong PR, sorry again!), #6693 introduced a second completions mechanism to fix a bug. Having two completion mechanisms isn't so nice. As @thufschmitt also pointed out, it was a bummer to go from `FlakeRef` to `std::string` when collecting flake refs. Now it is `FlakeRefs` again. The underlying issue that sought to work around was that completion of arguments not at the end can still benefit from the information from latter arguments. To fix this better, we rip out that change and simply defer all completion processing until after all the (regular, already-complete) arguments have been passed. In addition, I noticed the original completion logic used some global variables. I do not like global variables, because even if they save lines of code, they also obfuscate the architecture of the code. I got rid of them moved them to a new `RootArgs` class, which now has `parseCmdline` instead of `Args`. The idea is that we have many argument parsers from subcommands and what-not, but only one root args that owns the other per actual parsing invocation. The state that was global is now part of the root args instead. This did, admittedly, add a bunch of new code. And I do feel bad about that. So I went and added a lot of API docs to try to at least make the current state of things clear to the next person. -- This is needed for RFC 134 (tracking issue #7868). It was very hard to modularize `Installable` parsing when there were two completion arguments. I wouldn't go as far as to say it is *easy* now, but at least it is less hard (and the completions test finally passed). Co-authored-by: Valentin Gagarin <valentin.gagarin@tweag.io>
2023-10-23 15:03:11 +02:00
if (auto prefix = rootArgs.needsCompletion(*pos)) {
2020-05-10 20:32:21 +02:00
if (prefix == "-") {
Overhaul completions, redo #6693 (#8131) As I complained in https://github.com/NixOS/nix/pull/6784#issuecomment-1421777030 (a comment on the wrong PR, sorry again!), #6693 introduced a second completions mechanism to fix a bug. Having two completion mechanisms isn't so nice. As @thufschmitt also pointed out, it was a bummer to go from `FlakeRef` to `std::string` when collecting flake refs. Now it is `FlakeRefs` again. The underlying issue that sought to work around was that completion of arguments not at the end can still benefit from the information from latter arguments. To fix this better, we rip out that change and simply defer all completion processing until after all the (regular, already-complete) arguments have been passed. In addition, I noticed the original completion logic used some global variables. I do not like global variables, because even if they save lines of code, they also obfuscate the architecture of the code. I got rid of them moved them to a new `RootArgs` class, which now has `parseCmdline` instead of `Args`. The idea is that we have many argument parsers from subcommands and what-not, but only one root args that owns the other per actual parsing invocation. The state that was global is now part of the root args instead. This did, admittedly, add a bunch of new code. And I do feel bad about that. So I went and added a lot of API docs to try to at least make the current state of things clear to the next person. -- This is needed for RFC 134 (tracking issue #7868). It was very hard to modularize `Installable` parsing when there were two completion arguments. I wouldn't go as far as to say it is *easy* now, but at least it is less hard (and the completions test finally passed). Co-authored-by: Valentin Gagarin <valentin.gagarin@tweag.io>
2023-10-23 15:03:11 +02:00
rootArgs.completions->add("--");
for (auto & [flagName, flag] : shortFlags)
if (experimentalFeatureSettings.isEnabled(flag->experimentalFeature))
Overhaul completions, redo #6693 (#8131) As I complained in https://github.com/NixOS/nix/pull/6784#issuecomment-1421777030 (a comment on the wrong PR, sorry again!), #6693 introduced a second completions mechanism to fix a bug. Having two completion mechanisms isn't so nice. As @thufschmitt also pointed out, it was a bummer to go from `FlakeRef` to `std::string` when collecting flake refs. Now it is `FlakeRefs` again. The underlying issue that sought to work around was that completion of arguments not at the end can still benefit from the information from latter arguments. To fix this better, we rip out that change and simply defer all completion processing until after all the (regular, already-complete) arguments have been passed. In addition, I noticed the original completion logic used some global variables. I do not like global variables, because even if they save lines of code, they also obfuscate the architecture of the code. I got rid of them moved them to a new `RootArgs` class, which now has `parseCmdline` instead of `Args`. The idea is that we have many argument parsers from subcommands and what-not, but only one root args that owns the other per actual parsing invocation. The state that was global is now part of the root args instead. This did, admittedly, add a bunch of new code. And I do feel bad about that. So I went and added a lot of API docs to try to at least make the current state of things clear to the next person. -- This is needed for RFC 134 (tracking issue #7868). It was very hard to modularize `Installable` parsing when there were two completion arguments. I wouldn't go as far as to say it is *easy* now, but at least it is less hard (and the completions test finally passed). Co-authored-by: Valentin Gagarin <valentin.gagarin@tweag.io>
2023-10-23 15:03:11 +02:00
rootArgs.completions->add(std::string("-") + flagName, flag->description);
2020-05-10 20:32:21 +02:00
}
}
return false;
}
bool Args::processArgs(const Strings & args, bool finish)
{
if (expectedArgs.empty()) {
if (!args.empty())
throw UsageError("unexpected argument '%1%'", args.front());
return true;
}
Overhaul completions, redo #6693 (#8131) As I complained in https://github.com/NixOS/nix/pull/6784#issuecomment-1421777030 (a comment on the wrong PR, sorry again!), #6693 introduced a second completions mechanism to fix a bug. Having two completion mechanisms isn't so nice. As @thufschmitt also pointed out, it was a bummer to go from `FlakeRef` to `std::string` when collecting flake refs. Now it is `FlakeRefs` again. The underlying issue that sought to work around was that completion of arguments not at the end can still benefit from the information from latter arguments. To fix this better, we rip out that change and simply defer all completion processing until after all the (regular, already-complete) arguments have been passed. In addition, I noticed the original completion logic used some global variables. I do not like global variables, because even if they save lines of code, they also obfuscate the architecture of the code. I got rid of them moved them to a new `RootArgs` class, which now has `parseCmdline` instead of `Args`. The idea is that we have many argument parsers from subcommands and what-not, but only one root args that owns the other per actual parsing invocation. The state that was global is now part of the root args instead. This did, admittedly, add a bunch of new code. And I do feel bad about that. So I went and added a lot of API docs to try to at least make the current state of things clear to the next person. -- This is needed for RFC 134 (tracking issue #7868). It was very hard to modularize `Installable` parsing when there were two completion arguments. I wouldn't go as far as to say it is *easy* now, but at least it is less hard (and the completions test finally passed). Co-authored-by: Valentin Gagarin <valentin.gagarin@tweag.io>
2023-10-23 15:03:11 +02:00
auto & rootArgs = getRoot();
auto & exp = expectedArgs.front();
bool res = false;
2020-05-11 15:46:18 +02:00
if ((exp.handler.arity == ArityAny && finish) ||
(exp.handler.arity != ArityAny && args.size() == exp.handler.arity))
{
std::vector<std::string> ss;
Overhaul completions, redo #6693 (#8131) As I complained in https://github.com/NixOS/nix/pull/6784#issuecomment-1421777030 (a comment on the wrong PR, sorry again!), #6693 introduced a second completions mechanism to fix a bug. Having two completion mechanisms isn't so nice. As @thufschmitt also pointed out, it was a bummer to go from `FlakeRef` to `std::string` when collecting flake refs. Now it is `FlakeRefs` again. The underlying issue that sought to work around was that completion of arguments not at the end can still benefit from the information from latter arguments. To fix this better, we rip out that change and simply defer all completion processing until after all the (regular, already-complete) arguments have been passed. In addition, I noticed the original completion logic used some global variables. I do not like global variables, because even if they save lines of code, they also obfuscate the architecture of the code. I got rid of them moved them to a new `RootArgs` class, which now has `parseCmdline` instead of `Args`. The idea is that we have many argument parsers from subcommands and what-not, but only one root args that owns the other per actual parsing invocation. The state that was global is now part of the root args instead. This did, admittedly, add a bunch of new code. And I do feel bad about that. So I went and added a lot of API docs to try to at least make the current state of things clear to the next person. -- This is needed for RFC 134 (tracking issue #7868). It was very hard to modularize `Installable` parsing when there were two completion arguments. I wouldn't go as far as to say it is *easy* now, but at least it is less hard (and the completions test finally passed). Co-authored-by: Valentin Gagarin <valentin.gagarin@tweag.io>
2023-10-23 15:03:11 +02:00
bool anyCompleted = false;
2020-05-11 15:46:18 +02:00
for (const auto &[n, s] : enumerate(args)) {
Overhaul completions, redo #6693 (#8131) As I complained in https://github.com/NixOS/nix/pull/6784#issuecomment-1421777030 (a comment on the wrong PR, sorry again!), #6693 introduced a second completions mechanism to fix a bug. Having two completion mechanisms isn't so nice. As @thufschmitt also pointed out, it was a bummer to go from `FlakeRef` to `std::string` when collecting flake refs. Now it is `FlakeRefs` again. The underlying issue that sought to work around was that completion of arguments not at the end can still benefit from the information from latter arguments. To fix this better, we rip out that change and simply defer all completion processing until after all the (regular, already-complete) arguments have been passed. In addition, I noticed the original completion logic used some global variables. I do not like global variables, because even if they save lines of code, they also obfuscate the architecture of the code. I got rid of them moved them to a new `RootArgs` class, which now has `parseCmdline` instead of `Args`. The idea is that we have many argument parsers from subcommands and what-not, but only one root args that owns the other per actual parsing invocation. The state that was global is now part of the root args instead. This did, admittedly, add a bunch of new code. And I do feel bad about that. So I went and added a lot of API docs to try to at least make the current state of things clear to the next person. -- This is needed for RFC 134 (tracking issue #7868). It was very hard to modularize `Installable` parsing when there were two completion arguments. I wouldn't go as far as to say it is *easy* now, but at least it is less hard (and the completions test finally passed). Co-authored-by: Valentin Gagarin <valentin.gagarin@tweag.io>
2023-10-23 15:03:11 +02:00
if (auto prefix = rootArgs.needsCompletion(s)) {
anyCompleted = true;
ss.push_back(*prefix);
Overhaul completions, redo #6693 (#8131) As I complained in https://github.com/NixOS/nix/pull/6784#issuecomment-1421777030 (a comment on the wrong PR, sorry again!), #6693 introduced a second completions mechanism to fix a bug. Having two completion mechanisms isn't so nice. As @thufschmitt also pointed out, it was a bummer to go from `FlakeRef` to `std::string` when collecting flake refs. Now it is `FlakeRefs` again. The underlying issue that sought to work around was that completion of arguments not at the end can still benefit from the information from latter arguments. To fix this better, we rip out that change and simply defer all completion processing until after all the (regular, already-complete) arguments have been passed. In addition, I noticed the original completion logic used some global variables. I do not like global variables, because even if they save lines of code, they also obfuscate the architecture of the code. I got rid of them moved them to a new `RootArgs` class, which now has `parseCmdline` instead of `Args`. The idea is that we have many argument parsers from subcommands and what-not, but only one root args that owns the other per actual parsing invocation. The state that was global is now part of the root args instead. This did, admittedly, add a bunch of new code. And I do feel bad about that. So I went and added a lot of API docs to try to at least make the current state of things clear to the next person. -- This is needed for RFC 134 (tracking issue #7868). It was very hard to modularize `Installable` parsing when there were two completion arguments. I wouldn't go as far as to say it is *easy* now, but at least it is less hard (and the completions test finally passed). Co-authored-by: Valentin Gagarin <valentin.gagarin@tweag.io>
2023-10-23 15:03:11 +02:00
if (exp.completer) {
rootArgs.deferredCompletions.push_back({
.completer = exp.completer,
.n = n,
.prefix = *prefix,
});
}
} else
ss.push_back(s);
2020-05-11 15:46:18 +02:00
}
Overhaul completions, redo #6693 (#8131) As I complained in https://github.com/NixOS/nix/pull/6784#issuecomment-1421777030 (a comment on the wrong PR, sorry again!), #6693 introduced a second completions mechanism to fix a bug. Having two completion mechanisms isn't so nice. As @thufschmitt also pointed out, it was a bummer to go from `FlakeRef` to `std::string` when collecting flake refs. Now it is `FlakeRefs` again. The underlying issue that sought to work around was that completion of arguments not at the end can still benefit from the information from latter arguments. To fix this better, we rip out that change and simply defer all completion processing until after all the (regular, already-complete) arguments have been passed. In addition, I noticed the original completion logic used some global variables. I do not like global variables, because even if they save lines of code, they also obfuscate the architecture of the code. I got rid of them moved them to a new `RootArgs` class, which now has `parseCmdline` instead of `Args`. The idea is that we have many argument parsers from subcommands and what-not, but only one root args that owns the other per actual parsing invocation. The state that was global is now part of the root args instead. This did, admittedly, add a bunch of new code. And I do feel bad about that. So I went and added a lot of API docs to try to at least make the current state of things clear to the next person. -- This is needed for RFC 134 (tracking issue #7868). It was very hard to modularize `Installable` parsing when there were two completion arguments. I wouldn't go as far as to say it is *easy* now, but at least it is less hard (and the completions test finally passed). Co-authored-by: Valentin Gagarin <valentin.gagarin@tweag.io>
2023-10-23 15:03:11 +02:00
if (!anyCompleted)
exp.handler.fun(ss);
/* Move the list element to the processedArgs. This is almost the same as
`processedArgs.push_back(expectedArgs.front()); expectedArgs.pop_front()`,
except that it will only adjust the next and prev pointers of the list
elements, meaning the actual contents don't move in memory. This is
critical to prevent invalidating internal pointers! */
processedArgs.splice(
processedArgs.end(),
expectedArgs,
expectedArgs.begin(),
++expectedArgs.begin());
res = true;
}
if (finish && !expectedArgs.empty() && !expectedArgs.front().optional)
throw UsageError("more arguments are required");
return res;
}
nlohmann::json Args::toJSON()
{
auto flags = nlohmann::json::object();
for (auto & [name, flag] : longFlags) {
auto j = nlohmann::json::object();
j["hiddenCategory"] = hiddenCategories.count(flag->category) > 0;
if (flag->aliases.count(name)) continue;
if (flag->shortName)
j["shortName"] = std::string(1, flag->shortName);
if (flag->description != "")
j["description"] = trim(flag->description);
2021-01-25 19:03:13 +01:00
j["category"] = flag->category;
if (flag->handler.arity != ArityAny)
j["arity"] = flag->handler.arity;
if (!flag->labels.empty())
j["labels"] = flag->labels;
j["experimental-feature"] = flag->experimentalFeature;
flags[name] = std::move(j);
}
auto args = nlohmann::json::array();
for (auto & arg : expectedArgs) {
auto j = nlohmann::json::object();
j["label"] = arg.label;
j["optional"] = arg.optional;
if (arg.handler.arity != ArityAny)
j["arity"] = arg.handler.arity;
args.push_back(std::move(j));
}
auto res = nlohmann::json::object();
res["description"] = trim(description());
res["flags"] = std::move(flags);
res["args"] = std::move(args);
2020-12-07 13:04:24 +01:00
auto s = doc();
if (s != "") res.emplace("doc", stripIndentation(s));
return res;
}
Overhaul completions, redo #6693 (#8131) As I complained in https://github.com/NixOS/nix/pull/6784#issuecomment-1421777030 (a comment on the wrong PR, sorry again!), #6693 introduced a second completions mechanism to fix a bug. Having two completion mechanisms isn't so nice. As @thufschmitt also pointed out, it was a bummer to go from `FlakeRef` to `std::string` when collecting flake refs. Now it is `FlakeRefs` again. The underlying issue that sought to work around was that completion of arguments not at the end can still benefit from the information from latter arguments. To fix this better, we rip out that change and simply defer all completion processing until after all the (regular, already-complete) arguments have been passed. In addition, I noticed the original completion logic used some global variables. I do not like global variables, because even if they save lines of code, they also obfuscate the architecture of the code. I got rid of them moved them to a new `RootArgs` class, which now has `parseCmdline` instead of `Args`. The idea is that we have many argument parsers from subcommands and what-not, but only one root args that owns the other per actual parsing invocation. The state that was global is now part of the root args instead. This did, admittedly, add a bunch of new code. And I do feel bad about that. So I went and added a lot of API docs to try to at least make the current state of things clear to the next person. -- This is needed for RFC 134 (tracking issue #7868). It was very hard to modularize `Installable` parsing when there were two completion arguments. I wouldn't go as far as to say it is *easy* now, but at least it is less hard (and the completions test finally passed). Co-authored-by: Valentin Gagarin <valentin.gagarin@tweag.io>
2023-10-23 15:03:11 +02:00
static void hashTypeCompleter(AddCompletions & completions, size_t index, std::string_view prefix)
{
for (auto & type : hashTypes)
if (hasPrefix(type, prefix))
Overhaul completions, redo #6693 (#8131) As I complained in https://github.com/NixOS/nix/pull/6784#issuecomment-1421777030 (a comment on the wrong PR, sorry again!), #6693 introduced a second completions mechanism to fix a bug. Having two completion mechanisms isn't so nice. As @thufschmitt also pointed out, it was a bummer to go from `FlakeRef` to `std::string` when collecting flake refs. Now it is `FlakeRefs` again. The underlying issue that sought to work around was that completion of arguments not at the end can still benefit from the information from latter arguments. To fix this better, we rip out that change and simply defer all completion processing until after all the (regular, already-complete) arguments have been passed. In addition, I noticed the original completion logic used some global variables. I do not like global variables, because even if they save lines of code, they also obfuscate the architecture of the code. I got rid of them moved them to a new `RootArgs` class, which now has `parseCmdline` instead of `Args`. The idea is that we have many argument parsers from subcommands and what-not, but only one root args that owns the other per actual parsing invocation. The state that was global is now part of the root args instead. This did, admittedly, add a bunch of new code. And I do feel bad about that. So I went and added a lot of API docs to try to at least make the current state of things clear to the next person. -- This is needed for RFC 134 (tracking issue #7868). It was very hard to modularize `Installable` parsing when there were two completion arguments. I wouldn't go as far as to say it is *easy* now, but at least it is less hard (and the completions test finally passed). Co-authored-by: Valentin Gagarin <valentin.gagarin@tweag.io>
2023-10-23 15:03:11 +02:00
completions.add(type);
}
2020-05-04 22:40:19 +02:00
Args::Flag Args::Flag::mkHashTypeFlag(std::string && longName, HashType * ht)
{
2020-05-04 22:40:19 +02:00
return Flag {
.longName = std::move(longName),
.description = "hash algorithm ('md5', 'sha1', 'sha256', or 'sha512')",
.labels = {"hash-algo"},
.handler = {[ht](std::string s) {
*ht = parseHashType(s);
2020-05-10 21:50:32 +02:00
}},
Overhaul completions, redo #6693 (#8131) As I complained in https://github.com/NixOS/nix/pull/6784#issuecomment-1421777030 (a comment on the wrong PR, sorry again!), #6693 introduced a second completions mechanism to fix a bug. Having two completion mechanisms isn't so nice. As @thufschmitt also pointed out, it was a bummer to go from `FlakeRef` to `std::string` when collecting flake refs. Now it is `FlakeRefs` again. The underlying issue that sought to work around was that completion of arguments not at the end can still benefit from the information from latter arguments. To fix this better, we rip out that change and simply defer all completion processing until after all the (regular, already-complete) arguments have been passed. In addition, I noticed the original completion logic used some global variables. I do not like global variables, because even if they save lines of code, they also obfuscate the architecture of the code. I got rid of them moved them to a new `RootArgs` class, which now has `parseCmdline` instead of `Args`. The idea is that we have many argument parsers from subcommands and what-not, but only one root args that owns the other per actual parsing invocation. The state that was global is now part of the root args instead. This did, admittedly, add a bunch of new code. And I do feel bad about that. So I went and added a lot of API docs to try to at least make the current state of things clear to the next person. -- This is needed for RFC 134 (tracking issue #7868). It was very hard to modularize `Installable` parsing when there were two completion arguments. I wouldn't go as far as to say it is *easy* now, but at least it is less hard (and the completions test finally passed). Co-authored-by: Valentin Gagarin <valentin.gagarin@tweag.io>
2023-10-23 15:03:11 +02:00
.completer = hashTypeCompleter,
2020-05-04 22:40:19 +02:00
};
}
Args::Flag Args::Flag::mkHashTypeOptFlag(std::string && longName, std::optional<HashType> * oht)
{
return Flag {
.longName = std::move(longName),
.description = "hash algorithm ('md5', 'sha1', 'sha256', or 'sha512'). Optional as can also be gotten from SRI hash itself.",
.labels = {"hash-algo"},
.handler = {[oht](std::string s) {
*oht = std::optional<HashType> { parseHashType(s) };
}},
Overhaul completions, redo #6693 (#8131) As I complained in https://github.com/NixOS/nix/pull/6784#issuecomment-1421777030 (a comment on the wrong PR, sorry again!), #6693 introduced a second completions mechanism to fix a bug. Having two completion mechanisms isn't so nice. As @thufschmitt also pointed out, it was a bummer to go from `FlakeRef` to `std::string` when collecting flake refs. Now it is `FlakeRefs` again. The underlying issue that sought to work around was that completion of arguments not at the end can still benefit from the information from latter arguments. To fix this better, we rip out that change and simply defer all completion processing until after all the (regular, already-complete) arguments have been passed. In addition, I noticed the original completion logic used some global variables. I do not like global variables, because even if they save lines of code, they also obfuscate the architecture of the code. I got rid of them moved them to a new `RootArgs` class, which now has `parseCmdline` instead of `Args`. The idea is that we have many argument parsers from subcommands and what-not, but only one root args that owns the other per actual parsing invocation. The state that was global is now part of the root args instead. This did, admittedly, add a bunch of new code. And I do feel bad about that. So I went and added a lot of API docs to try to at least make the current state of things clear to the next person. -- This is needed for RFC 134 (tracking issue #7868). It was very hard to modularize `Installable` parsing when there were two completion arguments. I wouldn't go as far as to say it is *easy* now, but at least it is less hard (and the completions test finally passed). Co-authored-by: Valentin Gagarin <valentin.gagarin@tweag.io>
2023-10-23 15:03:11 +02:00
.completer = hashTypeCompleter,
2020-05-04 22:40:19 +02:00
};
}
Overhaul completions, redo #6693 (#8131) As I complained in https://github.com/NixOS/nix/pull/6784#issuecomment-1421777030 (a comment on the wrong PR, sorry again!), #6693 introduced a second completions mechanism to fix a bug. Having two completion mechanisms isn't so nice. As @thufschmitt also pointed out, it was a bummer to go from `FlakeRef` to `std::string` when collecting flake refs. Now it is `FlakeRefs` again. The underlying issue that sought to work around was that completion of arguments not at the end can still benefit from the information from latter arguments. To fix this better, we rip out that change and simply defer all completion processing until after all the (regular, already-complete) arguments have been passed. In addition, I noticed the original completion logic used some global variables. I do not like global variables, because even if they save lines of code, they also obfuscate the architecture of the code. I got rid of them moved them to a new `RootArgs` class, which now has `parseCmdline` instead of `Args`. The idea is that we have many argument parsers from subcommands and what-not, but only one root args that owns the other per actual parsing invocation. The state that was global is now part of the root args instead. This did, admittedly, add a bunch of new code. And I do feel bad about that. So I went and added a lot of API docs to try to at least make the current state of things clear to the next person. -- This is needed for RFC 134 (tracking issue #7868). It was very hard to modularize `Installable` parsing when there were two completion arguments. I wouldn't go as far as to say it is *easy* now, but at least it is less hard (and the completions test finally passed). Co-authored-by: Valentin Gagarin <valentin.gagarin@tweag.io>
2023-10-23 15:03:11 +02:00
static void _completePath(AddCompletions & completions, std::string_view prefix, bool onlyDirs)
2020-05-10 21:35:07 +02:00
{
Overhaul completions, redo #6693 (#8131) As I complained in https://github.com/NixOS/nix/pull/6784#issuecomment-1421777030 (a comment on the wrong PR, sorry again!), #6693 introduced a second completions mechanism to fix a bug. Having two completion mechanisms isn't so nice. As @thufschmitt also pointed out, it was a bummer to go from `FlakeRef` to `std::string` when collecting flake refs. Now it is `FlakeRefs` again. The underlying issue that sought to work around was that completion of arguments not at the end can still benefit from the information from latter arguments. To fix this better, we rip out that change and simply defer all completion processing until after all the (regular, already-complete) arguments have been passed. In addition, I noticed the original completion logic used some global variables. I do not like global variables, because even if they save lines of code, they also obfuscate the architecture of the code. I got rid of them moved them to a new `RootArgs` class, which now has `parseCmdline` instead of `Args`. The idea is that we have many argument parsers from subcommands and what-not, but only one root args that owns the other per actual parsing invocation. The state that was global is now part of the root args instead. This did, admittedly, add a bunch of new code. And I do feel bad about that. So I went and added a lot of API docs to try to at least make the current state of things clear to the next person. -- This is needed for RFC 134 (tracking issue #7868). It was very hard to modularize `Installable` parsing when there were two completion arguments. I wouldn't go as far as to say it is *easy* now, but at least it is less hard (and the completions test finally passed). Co-authored-by: Valentin Gagarin <valentin.gagarin@tweag.io>
2023-10-23 15:03:11 +02:00
completions.setType(Completions::Type::Filenames);
2020-05-11 15:46:18 +02:00
glob_t globbuf;
int flags = GLOB_NOESCAPE;
#ifdef GLOB_ONLYDIR
if (onlyDirs)
flags |= GLOB_ONLYDIR;
#endif
// using expandTilde here instead of GLOB_TILDE(_CHECK) so that ~<Tab> expands to /home/user/
if (glob((expandTilde(prefix) + "*").c_str(), flags, nullptr, &globbuf) == 0) {
for (size_t i = 0; i < globbuf.gl_pathc; ++i) {
if (onlyDirs) {
auto st = stat(globbuf.gl_pathv[i]);
if (!S_ISDIR(st.st_mode)) continue;
}
Overhaul completions, redo #6693 (#8131) As I complained in https://github.com/NixOS/nix/pull/6784#issuecomment-1421777030 (a comment on the wrong PR, sorry again!), #6693 introduced a second completions mechanism to fix a bug. Having two completion mechanisms isn't so nice. As @thufschmitt also pointed out, it was a bummer to go from `FlakeRef` to `std::string` when collecting flake refs. Now it is `FlakeRefs` again. The underlying issue that sought to work around was that completion of arguments not at the end can still benefit from the information from latter arguments. To fix this better, we rip out that change and simply defer all completion processing until after all the (regular, already-complete) arguments have been passed. In addition, I noticed the original completion logic used some global variables. I do not like global variables, because even if they save lines of code, they also obfuscate the architecture of the code. I got rid of them moved them to a new `RootArgs` class, which now has `parseCmdline` instead of `Args`. The idea is that we have many argument parsers from subcommands and what-not, but only one root args that owns the other per actual parsing invocation. The state that was global is now part of the root args instead. This did, admittedly, add a bunch of new code. And I do feel bad about that. So I went and added a lot of API docs to try to at least make the current state of things clear to the next person. -- This is needed for RFC 134 (tracking issue #7868). It was very hard to modularize `Installable` parsing when there were two completion arguments. I wouldn't go as far as to say it is *easy* now, but at least it is less hard (and the completions test finally passed). Co-authored-by: Valentin Gagarin <valentin.gagarin@tweag.io>
2023-10-23 15:03:11 +02:00
completions.add(globbuf.gl_pathv[i]);
}
2020-05-10 21:35:07 +02:00
}
globfree(&globbuf);
2020-05-10 21:35:07 +02:00
}
Overhaul completions, redo #6693 (#8131) As I complained in https://github.com/NixOS/nix/pull/6784#issuecomment-1421777030 (a comment on the wrong PR, sorry again!), #6693 introduced a second completions mechanism to fix a bug. Having two completion mechanisms isn't so nice. As @thufschmitt also pointed out, it was a bummer to go from `FlakeRef` to `std::string` when collecting flake refs. Now it is `FlakeRefs` again. The underlying issue that sought to work around was that completion of arguments not at the end can still benefit from the information from latter arguments. To fix this better, we rip out that change and simply defer all completion processing until after all the (regular, already-complete) arguments have been passed. In addition, I noticed the original completion logic used some global variables. I do not like global variables, because even if they save lines of code, they also obfuscate the architecture of the code. I got rid of them moved them to a new `RootArgs` class, which now has `parseCmdline` instead of `Args`. The idea is that we have many argument parsers from subcommands and what-not, but only one root args that owns the other per actual parsing invocation. The state that was global is now part of the root args instead. This did, admittedly, add a bunch of new code. And I do feel bad about that. So I went and added a lot of API docs to try to at least make the current state of things clear to the next person. -- This is needed for RFC 134 (tracking issue #7868). It was very hard to modularize `Installable` parsing when there were two completion arguments. I wouldn't go as far as to say it is *easy* now, but at least it is less hard (and the completions test finally passed). Co-authored-by: Valentin Gagarin <valentin.gagarin@tweag.io>
2023-10-23 15:03:11 +02:00
void Args::completePath(AddCompletions & completions, size_t, std::string_view prefix)
{
Overhaul completions, redo #6693 (#8131) As I complained in https://github.com/NixOS/nix/pull/6784#issuecomment-1421777030 (a comment on the wrong PR, sorry again!), #6693 introduced a second completions mechanism to fix a bug. Having two completion mechanisms isn't so nice. As @thufschmitt also pointed out, it was a bummer to go from `FlakeRef` to `std::string` when collecting flake refs. Now it is `FlakeRefs` again. The underlying issue that sought to work around was that completion of arguments not at the end can still benefit from the information from latter arguments. To fix this better, we rip out that change and simply defer all completion processing until after all the (regular, already-complete) arguments have been passed. In addition, I noticed the original completion logic used some global variables. I do not like global variables, because even if they save lines of code, they also obfuscate the architecture of the code. I got rid of them moved them to a new `RootArgs` class, which now has `parseCmdline` instead of `Args`. The idea is that we have many argument parsers from subcommands and what-not, but only one root args that owns the other per actual parsing invocation. The state that was global is now part of the root args instead. This did, admittedly, add a bunch of new code. And I do feel bad about that. So I went and added a lot of API docs to try to at least make the current state of things clear to the next person. -- This is needed for RFC 134 (tracking issue #7868). It was very hard to modularize `Installable` parsing when there were two completion arguments. I wouldn't go as far as to say it is *easy* now, but at least it is less hard (and the completions test finally passed). Co-authored-by: Valentin Gagarin <valentin.gagarin@tweag.io>
2023-10-23 15:03:11 +02:00
_completePath(completions, prefix, false);
}
Overhaul completions, redo #6693 (#8131) As I complained in https://github.com/NixOS/nix/pull/6784#issuecomment-1421777030 (a comment on the wrong PR, sorry again!), #6693 introduced a second completions mechanism to fix a bug. Having two completion mechanisms isn't so nice. As @thufschmitt also pointed out, it was a bummer to go from `FlakeRef` to `std::string` when collecting flake refs. Now it is `FlakeRefs` again. The underlying issue that sought to work around was that completion of arguments not at the end can still benefit from the information from latter arguments. To fix this better, we rip out that change and simply defer all completion processing until after all the (regular, already-complete) arguments have been passed. In addition, I noticed the original completion logic used some global variables. I do not like global variables, because even if they save lines of code, they also obfuscate the architecture of the code. I got rid of them moved them to a new `RootArgs` class, which now has `parseCmdline` instead of `Args`. The idea is that we have many argument parsers from subcommands and what-not, but only one root args that owns the other per actual parsing invocation. The state that was global is now part of the root args instead. This did, admittedly, add a bunch of new code. And I do feel bad about that. So I went and added a lot of API docs to try to at least make the current state of things clear to the next person. -- This is needed for RFC 134 (tracking issue #7868). It was very hard to modularize `Installable` parsing when there were two completion arguments. I wouldn't go as far as to say it is *easy* now, but at least it is less hard (and the completions test finally passed). Co-authored-by: Valentin Gagarin <valentin.gagarin@tweag.io>
2023-10-23 15:03:11 +02:00
void Args::completeDir(AddCompletions & completions, size_t, std::string_view prefix)
{
Overhaul completions, redo #6693 (#8131) As I complained in https://github.com/NixOS/nix/pull/6784#issuecomment-1421777030 (a comment on the wrong PR, sorry again!), #6693 introduced a second completions mechanism to fix a bug. Having two completion mechanisms isn't so nice. As @thufschmitt also pointed out, it was a bummer to go from `FlakeRef` to `std::string` when collecting flake refs. Now it is `FlakeRefs` again. The underlying issue that sought to work around was that completion of arguments not at the end can still benefit from the information from latter arguments. To fix this better, we rip out that change and simply defer all completion processing until after all the (regular, already-complete) arguments have been passed. In addition, I noticed the original completion logic used some global variables. I do not like global variables, because even if they save lines of code, they also obfuscate the architecture of the code. I got rid of them moved them to a new `RootArgs` class, which now has `parseCmdline` instead of `Args`. The idea is that we have many argument parsers from subcommands and what-not, but only one root args that owns the other per actual parsing invocation. The state that was global is now part of the root args instead. This did, admittedly, add a bunch of new code. And I do feel bad about that. So I went and added a lot of API docs to try to at least make the current state of things clear to the next person. -- This is needed for RFC 134 (tracking issue #7868). It was very hard to modularize `Installable` parsing when there were two completion arguments. I wouldn't go as far as to say it is *easy* now, but at least it is less hard (and the completions test finally passed). Co-authored-by: Valentin Gagarin <valentin.gagarin@tweag.io>
2023-10-23 15:03:11 +02:00
_completePath(completions, prefix, true);
}
Strings argvToStrings(int argc, char * * argv)
{
Strings args;
argc--; argv++;
while (argc--) args.push_back(*argv++);
return args;
}
std::optional<ExperimentalFeature> Command::experimentalFeature ()
{
return { Xp::NixCommand };
}
MultiCommand::MultiCommand(const Commands & commands_)
: commands(commands_)
{
2020-05-11 15:46:18 +02:00
expectArgs({
2020-08-17 19:33:18 +02:00
.label = "subcommand",
2020-05-11 15:46:18 +02:00
.optional = true,
2022-10-22 15:25:35 +02:00
.handler = {[=,this](std::string s) {
2020-05-11 15:46:18 +02:00
assert(!command);
auto i = commands.find(s);
if (i == commands.end()) {
std::set<std::string> commandNames;
for (auto & [name, _] : commands)
commandNames.insert(name);
auto suggestions = Suggestions::bestMatches(commandNames, s);
throw UsageError(suggestions, "'%s' is not a recognised command", s);
}
2020-05-11 15:46:18 +02:00
command = {s, i->second()};
command->second->parent = this;
}},
Overhaul completions, redo #6693 (#8131) As I complained in https://github.com/NixOS/nix/pull/6784#issuecomment-1421777030 (a comment on the wrong PR, sorry again!), #6693 introduced a second completions mechanism to fix a bug. Having two completion mechanisms isn't so nice. As @thufschmitt also pointed out, it was a bummer to go from `FlakeRef` to `std::string` when collecting flake refs. Now it is `FlakeRefs` again. The underlying issue that sought to work around was that completion of arguments not at the end can still benefit from the information from latter arguments. To fix this better, we rip out that change and simply defer all completion processing until after all the (regular, already-complete) arguments have been passed. In addition, I noticed the original completion logic used some global variables. I do not like global variables, because even if they save lines of code, they also obfuscate the architecture of the code. I got rid of them moved them to a new `RootArgs` class, which now has `parseCmdline` instead of `Args`. The idea is that we have many argument parsers from subcommands and what-not, but only one root args that owns the other per actual parsing invocation. The state that was global is now part of the root args instead. This did, admittedly, add a bunch of new code. And I do feel bad about that. So I went and added a lot of API docs to try to at least make the current state of things clear to the next person. -- This is needed for RFC 134 (tracking issue #7868). It was very hard to modularize `Installable` parsing when there were two completion arguments. I wouldn't go as far as to say it is *easy* now, but at least it is less hard (and the completions test finally passed). Co-authored-by: Valentin Gagarin <valentin.gagarin@tweag.io>
2023-10-23 15:03:11 +02:00
.completer = {[&](AddCompletions & completions, size_t, std::string_view prefix) {
for (auto & [name, command] : commands)
if (hasPrefix(name, prefix))
Overhaul completions, redo #6693 (#8131) As I complained in https://github.com/NixOS/nix/pull/6784#issuecomment-1421777030 (a comment on the wrong PR, sorry again!), #6693 introduced a second completions mechanism to fix a bug. Having two completion mechanisms isn't so nice. As @thufschmitt also pointed out, it was a bummer to go from `FlakeRef` to `std::string` when collecting flake refs. Now it is `FlakeRefs` again. The underlying issue that sought to work around was that completion of arguments not at the end can still benefit from the information from latter arguments. To fix this better, we rip out that change and simply defer all completion processing until after all the (regular, already-complete) arguments have been passed. In addition, I noticed the original completion logic used some global variables. I do not like global variables, because even if they save lines of code, they also obfuscate the architecture of the code. I got rid of them moved them to a new `RootArgs` class, which now has `parseCmdline` instead of `Args`. The idea is that we have many argument parsers from subcommands and what-not, but only one root args that owns the other per actual parsing invocation. The state that was global is now part of the root args instead. This did, admittedly, add a bunch of new code. And I do feel bad about that. So I went and added a lot of API docs to try to at least make the current state of things clear to the next person. -- This is needed for RFC 134 (tracking issue #7868). It was very hard to modularize `Installable` parsing when there were two completion arguments. I wouldn't go as far as to say it is *easy* now, but at least it is less hard (and the completions test finally passed). Co-authored-by: Valentin Gagarin <valentin.gagarin@tweag.io>
2023-10-23 15:03:11 +02:00
completions.add(name);
2020-05-11 15:46:18 +02:00
}}
});
2020-05-05 15:18:23 +02:00
categories[Command::catDefault] = "Available commands";
}
bool MultiCommand::processFlag(Strings::iterator & pos, Strings::iterator end)
{
if (Args::processFlag(pos, end)) return true;
2020-05-05 15:18:23 +02:00
if (command && command->second->processFlag(pos, end)) return true;
return false;
}
bool MultiCommand::processArgs(const Strings & args, bool finish)
{
if (command)
2020-05-05 15:18:23 +02:00
return command->second->processArgs(args, finish);
else
return Args::processArgs(args, finish);
}
nlohmann::json MultiCommand::toJSON()
{
auto cmds = nlohmann::json::object();
for (auto & [name, commandFun] : commands) {
auto command = commandFun();
auto j = command->toJSON();
2021-01-25 18:19:32 +01:00
auto cat = nlohmann::json::object();
cat["id"] = command->category();
cat["description"] = trim(categories[command->category()]);
cat["experimental-feature"] = command->experimentalFeature();
j["category"] = std::move(cat);
cmds[name] = std::move(j);
}
auto res = Args::toJSON();
res["commands"] = std::move(cmds);
return res;
}
}